Integrative machine and deep learning models for predictive analysis in complex disease areas

We aim to develop high-performance, explainable machine learning (ML) and deep learning (DL) models that can accurately predict severe outcomes in patients with complex diseases, such as stroke.

These models will leverage Sweden’s unique register data, integrate imaging, and generate new insights into optimal patient care.Stroke, a complex disease with diverse etiology and treatment approaches, serves as our primary case study.

In Sweden, approximately 23,000 strokes occur annually, with 25% of patients either deceased or dependent on assistance three months post-stroke.

To achieve our overall goal, we will first optimize and validate methods for the modelling framework, after that comprehensively assess the model on our wide range of data sources on disease outcomes, and finally interdisciplinary validate the models’ predictive capacity and usability.

For a group of patients who face a worse and more unpredictable prognosis, we will incorporate CT-images into our models.Our interdisciplinary team, comprising biostatisticians, ML/DL, and stroke specialists, has combined competence to develop, evaluate, and clinically validate the innovative models.

The knowledge generated will set the stage for research in predictive analysis in complex diseases at the same time offers opportunities that are crucial for identifying interventional targets at the individual, patient group, and healthcare management levels, thereby facilitating resource allocation for high-demand diseases.